A Novel Probabilistic-Based Deep Neural Network: Toward the Selection of Wart Treatment

Abstract

In clinical research, adequate use of gathered information to provide an intelligent framework to assist the doctors is a great challenge for the current biomedical research community. This study proposed a probabilistic deep neural network (PDNN) to select wart treatment method, where the layered structure of artificial neurons plays a crucial role in generating the optimal feature space. However, the probabilistic and thresholding technique is used to minimize the false negative and false positive instances. In the existing approaches, prediction accuracy and biasedness are major concerns in identifying the best wart treatment method. The benchmark dataset consists of 180 patients toward the selection of immunotherapy and cryotherapy treatment methods. Based on the feature descriptors about the wart, the baseline classifiers such as Naïve Bayes (NB), logistic regression and ensemble (LR), support vector machine (SVM), decision tree (DT), bagging, random forest (RF), and eXtreme Gradient Boosting (XGB) along with the developed PDNN was constructed by taking splitting ratio criteria into account. The standard statistical measures such as the measure of accuracy (MoA), error rate, sensitivity, specificity, and area under the curve (AUC) were considered to evaluate the predictive behavior. The proposed PDNN approach obtained promising results: moA, error rate, sensitivity, specificity, and measure of AUC as 0.9778, 0.0222, 0.9762, 0.9792, and 0.9818 while selecting immunotherapy and 0.9889, 0.0111, 1.0000, 0.9796, and 0.9970 in case of cryotherapy. The developed PDNN outperforms baseline classifiers and existing state-of-the-art wart treatment expert systems. The proposed model will improve the success rate and saves the diagnosing time. PDNN-based wart treatment identification system can be implemented in real time after consulting with a domain specialist.